Smalltalk › Squeak › Squeak - Dev

`2 raisedTo: 100000000` crashes VM

‹ Previous Topic Next Topic ›
 
classic Classic list List threaded Threaded
♦
♦
 12 messages Options
fniephaus
Reply | Threaded
Open this post in threaded view
♦
♦
|

`2 raisedTo: 100000000` crashes VM

fniephaus
Hi all,

We just found out that the VM crashes when trying to build very large integers.
I'm not sure what could cause the problem...

Best,
Fabio
Bert Freudenberg
Reply | Threaded
Open this post in threaded view
♦
♦
|

Re: `2 raisedTo: 100000000` crashes VM

Bert Freudenberg
On 01.06.2016, at 13:27, Fabio Niephaus <[hidden email]> wrote:
>
> Hi all,
>
> We just found out that the VM crashes when trying to build very large integers.
> I'm not sure what could cause the problem...

Maybe largeIntgrowTo doesn’t check the size?

Thread 0 Crashed:: Dispatch queue: com.apple.main-thread
0   libsystem_kernel.dylib         0x9b5fa572 __pthread_kill + 10
1   libsystem_pthread.dylib       0x91403654 pthread_kill + 101
2   libsystem_c.dylib             0x93dddc34 abort + 156
3   org.squeak.Squeak$(VM_MONIKER) 0x00139db9 sigsegv + 175
4   libsystem_platform.dylib       0x9e4bd79b _sigtramp + 43
5   ???                           0xffffffff 0 + 4294967295
6   org.squeak.Squeak$(VM_MONIKER) 0x00139d0a getCrashDumpFilenameInto + 82
7   org.squeak.Squeak$(VM_MONIKER) 0x00177262 largeIntgrowTo + 107
8   org.squeak.Squeak$(VM_MONIKER) 0x001771f1 normalizePositive + 190
9   org.squeak.Squeak$(VM_MONIKER) 0x0017678c primDigitMultiplyNegative + 597
10  ???                           0x04cb4cc8 0 + 80432328
11  org.squeak.Squeak$(VM_MONIKER) 0x000c203a interpret + 741


- Bert -






smime.p7s (5K) Download Attachment
Bert Freudenberg
Reply | Threaded
Open this post in threaded view
♦
♦
|

Re: 2 raisedTo: 100000000` crashes VM

Bert Freudenberg

> On 01.06.2016, at 13:43, Bert Freudenberg <[hidden email]> wrote:
>
> On 01.06.2016, at 13:27, Fabio Niephaus <[hidden email]> wrote:
>>
>> Hi all,
>>
>> We just found out that the VM crashes when trying to build very large integers.
>> I'm not sure what could cause the problem...
>
> Maybe largeIntgrowTo doesn’t check the size?
>
> Thread 0 Crashed:: Dispatch queue: com.apple.main-thread
> 0   libsystem_kernel.dylib         0x9b5fa572 __pthread_kill + 10
> 1   libsystem_pthread.dylib       0x91403654 pthread_kill + 101
> 2   libsystem_c.dylib             0x93dddc34 abort + 156
> 3   org.squeak.Squeak$(VM_MONIKER) 0x00139db9 sigsegv + 175
> 4   libsystem_platform.dylib       0x9e4bd79b _sigtramp + 43
> 5   ???                           0xffffffff 0 + 4294967295
> 6   org.squeak.Squeak$(VM_MONIKER) 0x00139d0a getCrashDumpFilenameInto + 82
> 7   org.squeak.Squeak$(VM_MONIKER) 0x00177262 largeIntgrowTo + 107
> 8   org.squeak.Squeak$(VM_MONIKER) 0x001771f1 normalizePositive + 190
> 9   org.squeak.Squeak$(VM_MONIKER) 0x0017678c primDigitMultiplyNegative + 597
> 10  ???                           0x04cb4cc8 0 + 80432328
> 11  org.squeak.Squeak$(VM_MONIKER) 0x000c203a interpret + 741
>
>
> - Bert -

Btw it worked fine in the interpreter and Cog.

- Bert -






smime.p7s (5K) Download Attachment
Eliot Miranda-2
Reply | Threaded
Open this post in threaded view
♦
♦
|

Re: 2 raisedTo: 100000000` crashes VM

Eliot Miranda-2


On Wed, Jun 1, 2016 at 5:29 AM, Bert Freudenberg <[hidden email]> wrote:

> On 01.06.2016, at 13:43, Bert Freudenberg <[hidden email]> wrote:
>
> On 01.06.2016, at 13:27, Fabio Niephaus <[hidden email]> wrote:
>>
>> Hi all,
>>
>> We just found out that the VM crashes when trying to build very large integers.
>> I'm not sure what could cause the problem...
>
> Maybe largeIntgrowTo doesn’t check the size?
>
> Thread 0 Crashed:: Dispatch queue: com.apple.main-thread
> 0   libsystem_kernel.dylib            0x9b5fa572 __pthread_kill + 10
> 1   libsystem_pthread.dylib           0x91403654 pthread_kill + 101
> 2   libsystem_c.dylib                 0x93dddc34 abort + 156
> 3   org.squeak.Squeak$(VM_MONIKER)    0x00139db9 sigsegv + 175
> 4   libsystem_platform.dylib          0x9e4bd79b _sigtramp + 43
> 5   ???                               0xffffffff 0 + 4294967295
> 6   org.squeak.Squeak$(VM_MONIKER)    0x00139d0a getCrashDumpFilenameInto + 82
> 7   org.squeak.Squeak$(VM_MONIKER)    0x00177262 largeIntgrowTo + 107
> 8   org.squeak.Squeak$(VM_MONIKER)    0x001771f1 normalizePositive + 190
> 9   org.squeak.Squeak$(VM_MONIKER)    0x0017678c primDigitMultiplyNegative + 597
> 10  ???                               0x04cb4cc8 0 + 80432328
> 11  org.squeak.Squeak$(VM_MONIKER)    0x000c203a interpret + 741
>
>
> - Bert -


Btw it worked fine in the interpreter and Cog.

So which VM does it fail on?
 

- Bert -









--
_,,,^..^,,,_
best, Eliot
Bert Freudenberg
Reply | Threaded
Open this post in threaded view
♦
♦
|

Re: 2 raisedTo: 100000000` crashes VM

Bert Freudenberg
On 01.06.2016, at 19:45, Eliot Miranda <[hidden email]> wrote:
> (2 raisedTo: 100000000) basicSize
> So which VM does it fail on?

Mac Spur 5.0.3692

- Bert -






smime.p7s (5K) Download Attachment
Denis Kudriashov
Reply | Threaded
Open this post in threaded view
♦
♦
|

Re: 2 raisedTo: 100000000` crashes VM

Denis Kudriashov
We have similar crash in Pharo by "200000 factorial"
crash.dmp includes last record #digitMultiply:neg:. It is same as here.

2016-06-01 22:48 GMT+02:00 Bert Freudenberg <[hidden email]>:
On 01.06.2016, at 19:45, Eliot Miranda <[hidden email]> wrote:
> (2 raisedTo: 100000000) basicSize
> So which VM does it fail on?

Mac Spur 5.0.3692

- Bert -
Eliot Miranda-2
Reply | Threaded
Open this post in threaded view
♦
♦
|

Re: 2 raisedTo: 100000000` crashes VM

Eliot Miranda-2
In reply to this post by Bert Freudenberg
Thanks all,

On Wed, Jun 1, 2016 at 1:48 PM, Bert Freudenberg <[hidden email]> wrote:
On 01.06.2016, at 19:45, Eliot Miranda <[hidden email]> wrote:
> (2 raisedTo: 100000000) basicSize
> So which VM does it fail on?

Mac Spur 5.0.3692

- Bert -

OK, tis fixed in VMMaker.oscog-eem.1879.

Reporting - 774 tallies, 6,836 msec.

**Tree**
...
[100 (6,836)  [] SmalltalkEditor(TextEditor) evaluateSelectionAndDo:
[  100 (6,836)  Compiler evaluate:in:to:notifying:ifFail:logged:
[    100 (6,836)  Compiler evaluateCue:ifFail:logged:
[      100 (6,836)  Compiler evaluateCue:ifFail:
[        100 (6,836)  UndefinedObject DoIt
[          100 (6,836)  AndreasSystemProfiler class spyOn:
[            100 (6,836)  BlockClosure ensure:
[              100 (6,836)  [] AndreasSystemProfiler class spyOn:
[                100 (6,836)  AndreasSystemProfiler spyOn:
[                  100 (6,836)  BlockClosure ensure:
[                    100 (6,836)  [] UndefinedObject DoIt
[                      100 (6,836)  SmallInteger(Integer) timesRepeat:
[                        100 (6,836)  [[]] UndefinedObject DoIt
[                          100 (6,836)  SmallInteger(Number) raisedTo:
[                            100 (6,836)  SmallInteger(Number) raisedToInteger:
[                              98.45 (6,730)  LargePositiveInteger *
[                                98.44 (6,730)  LargePositiveInteger(Integer) *
[                                  98.44 (6,730)  LargePositiveInteger(Integer) digitMultiply:neg:
[                                    50.36 (3,442)  SmalltalkImage primitiveGarbageCollect
[                                    28.3 (1,935)  LargePositiveInteger normalize
[                                      28.3 (1,935)  LargePositiveInteger(Integer) growto:
[                                        28.26 (1,932)  LargePositiveInteger(Integer) copyto:
[                                          28.26 (1,932)  SmalltalkImage primitiveGarbageCollect
**Leaves**
78.61 (5,374)  SmalltalkImage primitiveGarbageCollect

**Memory**
old -16,777,216 bytes
young +15,421,544 bytes
used -1,355,672 bytes
free -15,421,544 bytes

**GCs**
full 14 totalling 5,279ms (77.0% uptime), avg 377.0ms
incr 1 totalling 0ms (0.0% uptime), avg 0.0ms
tenures 0
root table 0 overflows

**Processes**
Total process switches: 1596
Without Profiler: 48
Stack page overflows: 4567
Stack page divorces: 15
 



--
_,,,^..^,,,_
best, Eliot
fniephaus
Reply | Threaded
Open this post in threaded view
♦
♦
|

Re: 2 raisedTo: 100000000` crashes VM

fniephaus
Thanks, Eliot!

On Thu, Jun 2, 2016 at 3:19 AM Eliot Miranda <[hidden email]> wrote:
Thanks all,

On Wed, Jun 1, 2016 at 1:48 PM, Bert Freudenberg <[hidden email]> wrote:
On 01.06.2016, at 19:45, Eliot Miranda <[hidden email]> wrote:
> (2 raisedTo: 100000000) basicSize
> So which VM does it fail on?

Mac Spur 5.0.3692

- Bert -

OK, tis fixed in VMMaker.oscog-eem.1879.

Reporting - 774 tallies, 6,836 msec.

**Tree**
...
[100 (6,836)  [] SmalltalkEditor(TextEditor) evaluateSelectionAndDo:
[  100 (6,836)  Compiler evaluate:in:to:notifying:ifFail:logged:
[    100 (6,836)  Compiler evaluateCue:ifFail:logged:
[      100 (6,836)  Compiler evaluateCue:ifFail:
[        100 (6,836)  UndefinedObject DoIt
[          100 (6,836)  AndreasSystemProfiler class spyOn:
[            100 (6,836)  BlockClosure ensure:
[              100 (6,836)  [] AndreasSystemProfiler class spyOn:
[                100 (6,836)  AndreasSystemProfiler spyOn:
[                  100 (6,836)  BlockClosure ensure:
[                    100 (6,836)  [] UndefinedObject DoIt
[                      100 (6,836)  SmallInteger(Integer) timesRepeat:
[                        100 (6,836)  [[]] UndefinedObject DoIt
[                          100 (6,836)  SmallInteger(Number) raisedTo:
[                            100 (6,836)  SmallInteger(Number) raisedToInteger:
[                              98.45 (6,730)  LargePositiveInteger *
[                                98.44 (6,730)  LargePositiveInteger(Integer) *
[                                  98.44 (6,730)  LargePositiveInteger(Integer) digitMultiply:neg:
[                                    50.36 (3,442)  SmalltalkImage primitiveGarbageCollect
[                                    28.3 (1,935)  LargePositiveInteger normalize
[                                      28.3 (1,935)  LargePositiveInteger(Integer) growto:
[                                        28.26 (1,932)  LargePositiveInteger(Integer) copyto:
[                                          28.26 (1,932)  SmalltalkImage primitiveGarbageCollect
**Leaves**
78.61 (5,374)  SmalltalkImage primitiveGarbageCollect

**Memory**
old -16,777,216 bytes
young +15,421,544 bytes
used -1,355,672 bytes
free -15,421,544 bytes

**GCs**
full 14 totalling 5,279ms (77.0% uptime), avg 377.0ms
incr 1 totalling 0ms (0.0% uptime), avg 0.0ms
tenures 0
root table 0 overflows

**Processes**
Total process switches: 1596
Without Profiler: 48
Stack page overflows: 4567
Stack page divorces: 15
 



--
_,,,^..^,,,_
best, Eliot
Bert Freudenberg
Reply | Threaded
Open this post in threaded view
♦
♦
|

Re: 2 raisedTo: 100000000` crashes VM

Bert Freudenberg
In reply to this post by Eliot Miranda-2
On 02.06.2016, at 03:19, Eliot Miranda <[hidden email]> wrote:
> OK, tis fixed in VMMaker.oscog-eem.1879.
>
> Reporting - 774 tallies, 6,836 msec.

Nice!

Is this finally a benchmark where SqueakJS can beat other VMs?

[10 timesRepeat: [(2 raisedTo: 100000000) basicSize]] timeToRun

Interp: 18042 ms
Cog: 7965 ms
Spur: ?
SqueakJS: 2821 ms (*)

:D

- Bert -

(*) http://bertfreudenberg.github.io/SqueakJS/run/#url=http://freudenbergs.de/bert/squeakjs&files=[Squeak4.5-13680.image,Squeak4.5-13680.changes,SqueakV41.sources]




smime.p7s (5K) Download Attachment
fniephaus
Reply | Threaded
Open this post in threaded view
♦
♦
|

Re: 2 raisedTo: 100000000` crashes VM

fniephaus
On my machine:

SqueakJS:  2832ms
RSqueak: 2523ms (*)

(*) using fallback Smalltalk code for large integer; https://www.hpi.uni-potsdam.de/hirschfeld/artefacts/rsqueak/bundle/RSqueak.zip


Maybe the JITs realize that the result of "2 raisedTo: ...." is not being used and omit the calculation?

--

On Thu, Jun 2, 2016 at 12:24 PM Bert Freudenberg <[hidden email]> wrote:
On 02.06.2016, at 03:19, Eliot Miranda <[hidden email]> wrote:
> OK, tis fixed in VMMaker.oscog-eem.1879.
>
> Reporting - 774 tallies, 6,836 msec.

Nice!

Is this finally a benchmark where SqueakJS can beat other VMs?

[10 timesRepeat: [(2 raisedTo: 100000000) basicSize]] timeToRun

Interp: 18042 ms
Cog: 7965 ms
Spur: ?
SqueakJS: 2821 ms (*)

:D

- Bert -

(*) http://bertfreudenberg.github.io/SqueakJS/run/#url=http://freudenbergs.de/bert/squeakjs&files=[Squeak4.5-13680.image,Squeak4.5-13680.changes,SqueakV41.sources]
timfelgentreff
Reply | Threaded
Open this post in threaded view
♦
♦
|

Re: 2 raisedTo: 100000000` crashes VM

timfelgentreff
I'm fairly sure that the RSqueak JIT realizes that with both 2 and 100000000 being constants in the doIt method, basic size is always going to return the same thing. In that case we're measuring the first ~1000 iterations, then the JIT time, and then only the cost of returning a constant.

Fabio Niephaus-3 wrote
On my machine:

SqueakJS:  2832ms
RSqueak: 2523ms (*)

(*) using fallback Smalltalk code for large integer;
https://www.hpi.uni-potsdam.de/hirschfeld/artefacts/rsqueak/bundle/RSqueak.zip


Maybe the JITs realize that the result of "2 raisedTo: ...." is not being
used and omit the calculation?

--

On Thu, Jun 2, 2016 at 12:24 PM Bert Freudenberg <[hidden email]>
wrote:

> On 02.06.2016, at 03:19, Eliot Miranda <[hidden email]> wrote:
> > OK, tis fixed in VMMaker.oscog-eem.1879.
> >
> > Reporting - 774 tallies, 6,836 msec.
>
> Nice!
>
> Is this finally a benchmark where SqueakJS can beat other VMs?
>
> [10 timesRepeat: [(2 raisedTo: 100000000) basicSize]] timeToRun
>
> Interp: 18042 ms
> Cog: 7965 ms
> Spur: ?
> SqueakJS: 2821 ms (*)
>
> :D
>
> - Bert -
>
> (*)
> http://bertfreudenberg.github.io/SqueakJS/run/#url=http://freudenbergs.de/bert/squeakjs&files=[Squeak4.5-13680.image,Squeak4.5-13680.changes,SqueakV41.sources]
>
>
>
Bert Freudenberg
Reply | Threaded
Open this post in threaded view
♦
♦
|

Re: 2 raisedTo: 100000000` crashes VM

Bert Freudenberg
One iteration of this “benchmark” is really just two dozen LargeInt multiplies (thanks to our logarithmic #raisedToInteger: implementation). The problem for most VMs is garbage, since the integers grow very fast very quickly. On the interpreter, 10 iterations do 580 (!) full GCs, whereas Cog “only” needs 30 full GCs. SqueakJS does 0 GCs.

- Bert -

> On 03.06.2016, at 14:25, timfelgentreff <[hidden email]> wrote:
>
> I'm fairly sure that the RSqueak JIT realizes that with both 2 and 100000000
> being constants in the doIt method, basic size is always going to return the
> same thing. In that case we're measuring the first ~1000 iterations, then
> the JIT time, and then only the cost of returning a constant.
>
>
> Fabio Niephaus-3 wrote
>> On my machine:
>>
>> SqueakJS:  2832ms
>> RSqueak: 2523ms (*)
>>
>> (*) using fallback Smalltalk code for large integer;
>> https://www.hpi.uni-potsdam.de/hirschfeld/artefacts/rsqueak/bundle/RSqueak.zip
>>
>>
>> Maybe the JITs realize that the result of "2 raisedTo: ...." is not being
>> used and omit the calculation?
>>
>> --
>>
>> On Thu, Jun 2, 2016 at 12:24 PM Bert Freudenberg &lt;
>
>> bert@
>
>> &gt;
>> wrote:
>>
>>> On 02.06.2016, at 03:19, Eliot Miranda &lt;
>
>> eliot.miranda@
>
>> &gt; wrote:
>>>> OK, tis fixed in VMMaker.oscog-eem.1879.
>>>>
>>>> Reporting - 774 tallies, 6,836 msec.
>>>
>>> Nice!
>>>
>>> Is this finally a benchmark where SqueakJS can beat other VMs?
>>>
>>> [10 timesRepeat: [(2 raisedTo: 100000000) basicSize]] timeToRun
>>>
>>> Interp: 18042 ms
>>> Cog: 7965 ms
>>> Spur: ?
>>> SqueakJS: 2821 ms (*)
>>>
>>> :D
>>>
>>> - Bert -
>>>
>>> (*)
>>> http://bertfreudenberg.github.io/SqueakJS/run/#url=http://freudenbergs.de/bert/squeakjs&files=[Squeak4.5-13680.image,Squeak4.5-13680.changes,SqueakV41.sources]
>>>
>>>
>>>
>
>
>
>
>
> --
> View this message in context: http://forum.world.st/2-raisedTo-100000000-crashes-VM-tp4898583p4899014.html
> Sent from the Squeak - Dev mailing list archive at Nabble.com.





smime.p7s (5K) Download Attachment
Free forum by Nabble Edit this page